Yes, I think we do. NIST have some conflicting
sounding statements, but they still have the one model (the partial floor model) in which the beams did NOT push the girder off its seat, and it still failed. This demonstrates that it could have failed. Certainly all the beam shear studs and the clip and seat bolts would have failed. This model was the the more detailed modelling of the connections than the larger 16-floor model.
And the limited expansion of the beams (which I accept for the sake of argument, but I'm not 100% convinced as to the precise limits) might not be as limiting as you think
The failures in the partial floor model is a complex three dimensional buckling distortion, not a simple linear push.
Consider this thought experiment
Now in the simple case, if K3004 were to push 4.6" then that's not far enough (6" needed) however there's at least a couple of scenarios where it might be, depending on the order of heating of the five beams (K, C, B, A, G) And the Girder (R, for convenience)
Take a purely imaginary outside case. K expands first (roughly), breaking the C79-R connections. C, B, A, do likewise. Now if we (for now) ignore K,C, B, A, and imagine for the sake of argument that the R pivots around C44.
Under this very simplistic model an expansion one inch of G would result in motion of the end of the girder at C78 by of approximately eight inches.
Now that's not realistic. However nor do I think a simplistic model of KCBAG all expanding at the same rate is realistic. They would have expanded at different rates, and the combination of those expansions, with perhaps some pivoting, and distortion of the girder itself, would be what caused the girder to be unseated on C79.
NIST describes that the C79 connection was unseated by the girder walking off. They ascribe this to the five beams expanding. They might not be wrong, and they do not (I believe) claim that it was simply due to the expansion of K3004.
Ideally we'd like to see the graph of the positions of the endpoints of all the beams and girders involved. Unfortunately that's unlikely. However the end result still seems plausible.